Ophthalmic medical image segmentation method and system and storage medium

The present application claims the priority of Chinese patent application No. 202411254318.2, filed on 2024 Sep. 9, the entire disclose of which is incorporated herein by reference.

The present invention relates to the technical field of medical image segmentation, in particular to an ophthalmic medical image segmentation method and system and a storage medium.

Medical image segmentation can make images of pathological structures clearer, and thus play an important role in computer-aided diagnosis and intelligent medical treatment.

The main difficulties in medical image segmentation for ophthalmic diseases are as follows: morphological features presented by a lesion region are diverse, with uneven size, shape and intensity distribution; and the boundary between the lesion region and surrounding tissues is unclear, possibly accompanied by complications of patients, such as effusion. These difficulties increase the complexity of medical image segmentation, and the results obtained by the existing ophthalmic medical image segmentation methods often contain a large number of relevant redundant information and irrelevant feature information, resulting in low segmentation progress and low efficiency.

In addition, the existing ophthalmic medical image segmentation methods use a single loss function to construct convolutional neural networks, which perform poorly when processing special samples, lack comprehensiveness, and have low segmentation accuracy.

For this purpose, the technical problem to be solved by the present invention is to overcome the problem that an ophthalmic medical image segmentation method in the prior art is low in segmentation efficiency and accuracy and lacks comprehensiveness, and to provide an ophthalmic medical image segmentation method and system and a storage medium, which achieve fast segmentation speed and high efficiency, and can obtain a high-accuracy segmentation result while possessing comprehensiveness.

In a first aspect, to solve the above technical problem, the present invention provides an ophthalmic medical image segmentation method. The method includes the following steps:

In one embodiment of the present invention, a method of splicing the Iand the Iis as follows:

In one embodiment of the present invention, the weighted loss function is constructed using a multi-loss fusion method, which is as follows:

In one embodiment of the present invention, the multi-classified logistic loss function L(Y,Ŷ) is as follows:

In one embodiment of the present invention, the encoder is configured with a plurality of layers, and each layer is subjected to the convolution operation, a batch normalization operation and a maximum pooling operation; and the size of a convolution kernel and the number of operations for each layer to perform the convolution operation can be set autonomously.

In one embodiment of the present invention, the encoder is configured with four layers, each layer is subjected to the convolution operation twice based on a convolution layer having a convolution kernel size of 3*3 and a step size of 1, and each layer is subjected to the maximum pooling operation based on a pooling layer having a pooling kernel size of 2*2.

In one embodiment of the present invention, the medical image data includes OCT image data of ophthalmic choroidal neovascularization and fundus color image data of glaucoma, and the OCT image data and the fundus color image data both contain original image data and corresponding gold standard image data.

In one embodiment of the present invention, background pixel values in the OCT image data and the fundus color image data are assigned to 0, and the same pixel values are assigned to the lesion regions of the respective categories in the corresponding gold standard image data and incremented sequentially according to their categories.

In a second aspect, to solve the above technical problem, the present invention further provides an ophthalmic medical image segmentation system. The system includes:

In a third aspect, to solve the above technical problem, the present invention further provides a computer-readable storage medium having stored a computer program therein, the computer program, when executed by a processor, implementing the steps of the ophthalmic medical image segmentation method.

Compared with the prior art, the above technical solutions of the present invention have the following beneficial effects.

According to the ophthalmic medical image segmentation method and system and the storage medium provided by the present invention, the convolutional neural network model adopting the U-shaped encoding and decoding structure is constructed based on the attention mechanism and the weighted loss function, such that the segmentation progress of medical images is fast and efficient; special samples can be effectively processed, achieving comprehensiveness; and the segmentation results are highly accurate.

It is to be understood that the specific embodiments described herein are only used for explaining the present application, and are not used for limiting the present application.

The technical solutions in the embodiments of the present application will be described clearly and completely below in conjunction with accompanying drawings in the embodiments of the present application. Of course, the described embodiments are merely some embodiments, rather than all embodiments, of the present application.

Medical image segmentation for ophthalmic diseases needs to be achieved in combination with disease characteristics. For example, choroidal neovascularization (CNV) is a typical symptom of age-related macular degeneration. The rupture of the neovascularization of a choroidal layer in a retinal layer will cause decreased vision or even blindness in patients. Glaucoma is characterized by an increase in areas of an optic cup and an optic disc. In addition, optical coherence tomography (OCT) is a high-resolution non-invasive imaging technology that may record and display various structures of the fundus.

Efficient and accurate medical image segmentation of OCT images accompanied by CNV and glaucoma fundus color images is exactly the invention objective that the inventors of embodiments of the present application intend to achieve. Therefore, the embodiments of the present application provide an ophthalmic medical image segmentation method and system and a storage medium.

The present embodiment provides an ophthalmic medical image segmentation method. As shown in, the method includes the following steps:

According to the ophthalmic medical image segmentation method provided by the present embodiment, the convolutional neural network model adopting the U-shaped encoding and decoding structure is constructed, the attention mechanism is introduced in the U-shaped encoding and encoding structure, and the convolutional neural network model adopts the weighted loss function, such that (1) the segmentation progress of medical images is fast and efficient; (2) special samples can be effectively processed, achieving comprehensiveness; and (3) the segmentation results of the medical images are highly accurate.

Next, the ophthalmic medical image segmentation method provided by the present embodiment will be described in detail.

I. Principle of Method

Step 1:

Optionally, medical images of the ophthalmic lesion region are labeled, the medical image data is acquired, and the medical image data is divided into a training set and a test set according to a proportion of 8:2.

Optionally, the medical image data includes OCT image data of ophthalmic choroidal neovascularization and fundus color image data of glaucoma, and the OCT image data and the fundus color image data both contain original image data and corresponding gold standard image data, wherein the gold standard image represents segmented regions manually labeled by a professional doctor or under the guidance of a professional doctor.

Optionally, background pixel values in the OCT image data and the fundus color image data are assigned to 0, and the same pixel values are assigned to the lesion regions of the respective categories in the corresponding gold standard image data and incremented sequentially according to their categories.

Step 2:

Optionally, a method of splicing the Iand the Iis as follows:

Specifically, the encoder is configured with a plurality of layers, and each layer is subjected to the convolution operation, a batch normalization operation and a maximum pooling operation; and the size of a convolution kernel and the number of operations for each layer to perform the convolution operation can be set autonomously.

Specifically, the increase in the number of layers of the encoder can capture more feature levels, but will increase the computational cost and memory requirements and increase the training difficulty; the decrease in the number of layers of the encoder will usually make a constructed network model simpler and reduce the amount of computation, but also reduce the expressive ability of the constructed network model, making it difficult to capture detailed features.

Preferably, the encoder is configured with four layers, each layer is subjected to the convolution operation twice based on a convolution layer having a convolution kernel size of 3*3 and a step size of 1, and each layer is subjected to the maximum pooling operation based on a pooling layer having a pooling kernel size of 2*2.

Exemplarily, as shown in, by taking an OCT image of 512×512CNV as an example, this image is input into the left encoder section, and each layer of the encoder performs the convolution operation twice, and then performs the maximum pooling operation to reduce a spatial resolution; the number of channels of a convolutional feature map gradually increases from 64 at the beginning to 128, 256, 512 and 1024 sequentially after passing through the encoder layer by layer; and the right decoder section corresponds to the left encoder section, deep layer features in each layer of the decoder are subjected to the up-sampling operation and then integrated together with low layer features of the corresponding encoder section into the attention module A, and the number of channels of the convolutional feature map is adjusted sequentially to 512, 256, 128, and 2.

In, Conv3×3 represents that the convolution operation is performed based on a convolution layer having a convolution kernel size of 3*3; ReLU represents the activation function; Copy and crop represents that the splicing operation is performed; Max pool2×2 represents that the maximum pooling operation is performed based on a pooling layer having a pooling kernel size of 2*2; Up-Conv2×2 represents that the up-sampling operation of 2*2 is performed; and Attention Module represents the attention module A.

Specifically, in the above example, rich information of the superficial layer features of the encoder is transmitted to a deep network, thereby improving the overall performance of the convolutional neural network model.

Specifically, the U-shaped encoding and decoding structure can effectively focus the lesion region, suppress interference caused by other irrelevant regions, and improve the segmentation accuracy of the lesion region.

Specifically, the weighted loss function is constructed using a multi-loss fusion method, which is as follows:

Specifically, the multi-classified logistic loss function L(Y,Ŷ) is as follows: